Table 3.

Performance for each predictive model on the CIRCLE-seq data set.

Metric	CRISPR-IP	FNN3	FNN5	FNN10	CNN3	CNN5	LSTM	GRU	Encoding
Accuracy	0.990	0.670	0.796	0.955	0.982	0.982	0.988	0.987	encoding scheme 1
Accuracy	0.990	0.971	0.966	0.984	0.990	0.989	0.988	0.988	encoding scheme 2
F1 score	0.621	0.144	0.133	0.364	0.222	0.255	0.560	0.518	encoding scheme 1
F1 score	0.644	0.364	0.403	0.472	0.533	0.501	0.569	0.531	encoding scheme 2
PR-AUC	0.695	0.209	0.082	0.302	0.316	0.319	0.665	0.616	encoding scheme 1
PR-AUC	0.751	0.350	0.339	0.391	0.641	0.587	0.682	0.676	encoding scheme 2
Precision	0.808	0.117	0.087	0.325	0.691	0.672	0.669	0.688	encoding scheme 1
Precision	0.791	0.529	0.439	0.538	0.882	0.887	0.725	0.796	encoding scheme 2
ROC-AUC	0.973	0.767	0.775	0.855	0.891	0.885	0.970	0.965	encoding scheme 1
ROC-AUC	0.982	0.892	0.900	0.904	0.968	0.961	0.971	0.973	encoding scheme 2
Recall	0.526	0.711	0.614	0.566	0.194	0.211	0.569	0.504	encoding scheme 1
Recall	0.593	0.473	0.583	0.555	0.396	0.364	0.570	0.506	encoding scheme 2

Notes: Better results are indicated in bold. Encoding scheme 1 was proposed by Lin et al., and coding scheme 2 was proposed by us.